ANGIOSTATIN Generation, Structure and Function of the Isoforms

ANGIOSTATIN

Generation, Structure and Function of the Isoforms

Jennifer A. Doll and Gerald A. Soff

Northwestern University Feinberg School of Medicine, Department of Medicine, Division oj Hernatology/Oncology and the Robert H. Lurie Comprehensive Cancer Center, Chicago, ZL

1. INTRODUCTION: DISCOVERY AND FUNCTION OF ANGIOSTATIN

The discovery of angiostatin in 1994 provided a major advance in the field of angiogenesis research. As Folkman first proposed in 1971, angiogenesis, the growth of new vessels from the pre-existing vasculature, is required for tumors to grow beyond a few millimeters in diameter and for tumor metastasis'. With a few exceptions (wound healing and reproductive cycles), the vasculature in the adult is maintained in a quiescent state by a net balance of angiogenic inducers and inhibitors secreted into the tissue Tumors shift this balance to favor vessel growth, by increasing inducer levels, decreasing inhibitor levels, or most often, a combination of Tumor angiogenesis involves many processes, including increased vascular permeability, endothelial cell activation, proliferation, migration and tube formation as well as matrix degradation. Designing therapies targeting any of these steps would inhibit angiogenesis, and thus inhibit tumor growth. Therefore, much research has been devoted to developing such agents for use in cancer therapy.

Numerous inducers and inhibitors of angiogenesis have now been identified, both endogenous and exogenous. It is of interest to note that the endogenous regulators, both inducers and inhibitors, span extremely diverse groups of molecules, including growth factors and cytokines, proteins and enzymes, protein cleavage products, enzyme inhibitors, carbohydrates, lipids, hormones and Most of the naturally occurring angiogenesis inhibitors, such as thrombospondin-1 and pigment epithelium-

176 CYTOKINES AND CANCER derived factor, have a wide variety of cellular activities and affect multiple cell types6,'. Few inhibitors have been identified which have high specificity to activated endothelial cell suppression. Angiostatin was the first natural inhibitor discovered and one of the few to show high selectivity for the endothelial cells lining the blood

Angiostatin's discovery stemmed from research into several observations that had perplexed clinicians and researchers alike for many years. First, as described in some case reports as well as in animal tumor models, rapid growth of distant metastases has been observed following the removal of the primary tumor (reviewed in 8). The second observation is that a secondary tumor can be suppressed by the presence of a different primary tumor at a distant location (reviewed in 8). From these observations and their knowledge of tumor angiogenesis, Folkman developed the hypothesis that some tumors, while able to stimulate angiogenesis within their own capillary beds, produce angiogenesis inhibitors which enter the circulation and suppress angiogenesis in metastatic foci8. They tested this hypothesis using a variant of the murine Lewis lung carcinoma (LLC) cell line with a low metastatic potential (LLC-LM)~. They resected the primary subcutaneous LLC-LM tumors 14 days after implantation in mice and compared metastatic growth to sham operated mice in which the primary tumor was left intact8.

Mice with resected primary tumors had 10-fold more metastatic growth compared to sham-operated mice, suggesting that the primary tumor had been inhibiting the growth of metastases8. In addition, corneal neovascularization toward an implanted pellet containing basic fibroblast growth factor (bFGF), a potent angiogenesis inducer, was inhibited in mice with intact primary tumors but not in mice with resected tumors, indicating that a circulating factor was indeed inhibiting angiogenesis8

From more than 100 liters of urine collected from mice bearing LLC-LM tumors, O'Reilly and colleagues isolated a 38 kDa murine protein, which they named angiostatin8. By sequence analysis, they determined that this protein was an internal fragment of plasminogen (PLG), beginning with amino acid 98 (initial sequence of 98-102: valine-tyrosine-leucine-serine- glutamic acid) and with a C-terminus at approximately amino acid 4408.

This fragment included the first four of five loop structures, called kringle domains, in the PLG protein (Figure l)8. The angiostatin produced in the mice was dependent on the presence of the Lewis lung tumor8. However, at this time, it was not known if the tumor itself was producing the angiostatin or if the tumor was producing protein(s) that could generate angiostatin or that could block inhibitors of PLG activators8.

To generate human angiostatin for study, O'Reilly and colleagues digested human PLG with elastase, as it was known to liberate kringle

containing fragments, including a fragment comprised of kringle domains 1- 3, isolated kringle 4 and mini-PLG, which is kringle 5 attached to the plasmin catalyhc domain". From the elastase digestion, they isolated a fragment of approximately 40 kl3as. This fragment included the first three

Figure I. Structure of Human Plasminogen and Angiostatin K1-4. A) Plasminogen, the zyrnogen form of plasmin, contains five conserved kringle domains (K1 - KS), as well as

the protease domain. The triangle indicates where plasminogen activators (uPA, tPA) cleave plasminogen to yield the active serine protease plasmin (picture thanks to M. Llinas and co-workers, Carnegie Mellon University, Pittsburgh, PA). B) Angiostatin as originally

described by O'Reilly et aL8, consists of the first four of the five kringle domains of plasminogen.

kringle domains of human PLG, with an N-terminus of amino acid 97 or 99 of the human PLG protein, a region corresponding to the murine angiostatin8. The purified elastase-generated human angiostatin specifically

178 CYTOKINES AND CANCER

inhibited endothelial cell proliferation in vitro and inhibited neovascularization in the in vivo chick chorioallantoic membrane (CAM) assay8. In the LLC-LM model, following removal of the primary tumor, systemic treatment with the elastase-generated human angiostatin suppressed metastases8. Elastase-generated human angiostatin also inhibited bFGF and vascular endothelial growth factor (VEGF)-induced endothelial cell migration and tube formation in a collagen matrix system". By careful histologic analysis, this group found that dormancy of micro metastases in mice with intact LLC tumors was due to a balance between proliferation and apoptosis of the tumor cells12. When the primary tumor was removed, and thus the circulating angiostatin also removed, angiogenesis ensued and apoptosis was significantly decreased, allowing for expansion of the metastatic foci12. A later study showed that the elastase-generated human angiostatin could also inhibit primary tumor growth of human prostate, breast and colon cancer cells in subcutaneous mouse models treated with a systemic dose of 50 mgkg twice daily9. Consistent with their findings in the LLC model, this suppression was also due to an increase in apoptotic rate while the proliferation rate remained unchanged9.

In another study, Folkman and colleagues generated an expression vector to produce recombinant murine angiostatin, and the purified protein encompassed the first four kringle domains with an amino acid sequence as follows: AspZO through Ser32-Ser-Arg97 through Gly4,, 13 . This recombinant angiostatin was significantly larger than the in vivo generated angiostatin, at 52 kDa versus 38 kDa13. The N-terminal addition of 14 amino acids contributed to this increase but could not account for the total difference in size; therefore, the remaining difference was thought to be due to glycosylation differenced3. The size difference did not affect activity, and in fact, the recombinantly generated angiostatin was more potent than the elastase-generated human angiostatin against endothelial proliferation in vitro and inhibited LLC-LM subcutaneous primary tumor growth in vivo13.

They further showed that when T241 fibrosarcoma tumor cells were transfected with an angiostatin expression vector and implanted in mice, primary and metastatic tumor growth were both inhibited14.

Subsequently, other researchers isolated angiostatin and angiostatin-like proteins. These related proteins, or angiostatin isoforms, had differing NH2- and COOH-termini of PLG and varied mainly in their kringle domain content. The differences in structure and in anti-endothelial cell and anti- tumor activity are discussed in detail below. Overall, studies by several groups, including our own, confirmed that angiostatin isoforms inhibit endothelial cell proliferation, migration and tube formation induced by a variety of angiogenesis inducers in ~ i t r o " - ' ~ and also inhibit vessel formation

in vivo in the corneal pocket assay, in the embryonic body model and in the aortic ring m ~ d e l ' ~ " ~ " ~ . The anti-tumor activity of angiostatin isoforms has now been demonstrated against a variety of tumor types in mouse models as well, including hemangioendotheliomaZ0, glioma21~22, liver c a n ~ e r ~ ~ - ~ ' , lung c a n ~ e r ~ ~ ' ~ ' , ovarian cancerz8, colorectal ~ a n c e ? ~ and breast ~ancer'~>~O. As angiogenesis is important in physiologic and pathophysiologic processes in addition to cancer, angiostatin isoforms have been investigated in other diseases. Potential therapeutic benefits have also been observed in models of corneal wound healing3', collagen-induced arthritis32133 and endometri~sis~~.

In contrast to angiostatin, the parent PLG protein does not affect endothelial cell proliferation in vitro8. PLG is a 92 kDa zymogen produced in the liver, although other cells may be capable of producing it, such as eosinophils, kidney and corneal cells35. In the adult, the PLG plasma concentration is -200 mg/L, a concentration equivalent to -2 pM3'. FLG is activated to form the serine protease plasmin by endogenous PLG activators (PA), tissue-type PA (tPA) or urokinase-type PA (uPA), which cleave PLG between Arg561 and Va1562, with the complete protein remaining intact, linked by two disulfide bonds35. The PLG / plasmin protein consists of 791 amino acids, with an 0-linked glycosylation at Thr346 and an N-linked glycosylation at ~ s n 2 8 9 ~ ' . The main function of the PAlplasmin system is fibrinolysis via fibrin degradation3'. Plasmin also degrades the extracellular matrix (ECM) by degradation of fibronectin, laminin and type IV collagen and indirectly through activation of matrix metalloproteinases (MMP) (reviewed in 36,37). It can also activate and/or stimulate growth factor release from the ECM, including VEGF, bFGF, hepatocyte growth factor (HGF) and transforming growth factor beta (reviewed in 36). Thus, this system affects cell adhesion, cell migration and cell-to-cell signaling.

However, while angiostatin is clearly anti-angiogenic and tumor suppressive, the role of the PA / plasmin system in tumor growth remains unclear. Some studies suggest that activation of the plasmin system promotes tumorigenesis while others suggest that inhibition of this pathway promotes tumorigenesis.

Increased levels of the PLG activator uPA and its cell surface receptor, uPAR, are a negative prognostic factor for several cancer types, including breast, gastric, colon, lung, prostate and ovarian cancers (reviewed in 36).

tPA is also produced by some melanoma and neuroblastoma t ~ m o r s ~ ~ - ~ ' . Increased uPA, uPAR or tPA could increase plasmin activity. Thus, tumorigenesis and/or angiogenesis could be stimulated through release of growth factors and/or activation of MMPs. A recent study demonstrated that plasmin can activate VEGF isoforms C and D, which stimulate lymphatic angiogenesis and vascular angiogenesis, respectively42. Several studies also suggest that plasmin can directly stimulate bovine aortic endothelial cell migration in vitro43144, and it can also induce bovine capillary endothelial cell

180 CYTOKINES AND CANCER proliferation and spreading via induction of stress fiber formation, while PLG had no effect44.

Contrasting studies show that increased levels of plasminogen activator inhibitor-1 (PAI-I), an inhibitor of PA, correlates with poor prognosis in several cancer types, including breast, ovarian and cervical cancers and gliomas (reviewed in 36,45,46). In addition, we have shown that plasmin activation is necessary for angiostatin generation on the cancer cell surface, production of which would potentially be inhibit01-y~~. We have also shown that an angiostatin isoform can inhibit plasmin-induced migration of bovine aortic endothelial cells43. These data suggest that the role of the PA 1 plasmin system in cancer requires further study. We may find that the pro- or anti-tumor activity of this system depends on the cell type and/or on the relative expression levels of PA versus inhibitors and/or on the level of angiostatin production. The level of angiostatin generation may depend, in part, on factors outside the PA / plasmin system. For example, we recently demonstrated that PC-3 prostate cancer cells generate an angiostatin isoform on the cell surface, dependent on the presence of uPA and cell surface P- a ~ t i n ~ ~ ' ~ ~ . Therefore, cells, that have increased plasmin activity due to increased uPA / uPAR expression but also have p-actin expression (or another angiostatin-generating cofactor), may have increased angiostatin levels which could suppress tumor growth and metastases. In tumors with increased uPA / uPAR expression but with no p-actin (or other cofactor), tumor progression may be facilitated by the increased plasmin activity.

Ultimately, the plasmin activity levels depend on the balance of activators and inhibitors present. Therefore, to determine the role of this system in different cancer types, we will likely need to assess relative expression levels of each component of the system and/or measure plasmin activity levels and angiostatin production levels.

2. GENERATION OF ANGIOSTATIN ISOFORMS

The ability of elastase digestion of PLG to generate kringle-containing fragments has been known for decades1'. However, it has become very evident that a variety of mechanisms can be used to produce angiostatin isoforms, which vary in their N- and COOH-termini and in the number of kringle domains they contain. As it has also become clear that the exact kringle content of each isoform is critical to its activity, for clarity and ease of discussion, we will designate the different isoforms by their kringle domain (K) content henceforward. Using this designation system, the original murine angiostatin isolated from LLC-LM tumor bearing mice is

angiostatin K1-4, and the human angiostatin generated by elastase digestion is angiostatin K1-3. As discussed above, elastase digestion of human PLG is known to liberate K1-3, isolated K4 and mini-PLG". A modified version of this digestion has also been used to generate ~ 1 - 4 " and isolated ~ 5 " . However, the relevance of elastase digestion to in vivo generation mechanisms is not clear.

Folkman's group later found that the angiostatin K1-4 isolated from tumor bearing mice was generated via PLG cleavage by MMP-2 (gelatinase A) released by the LLC-LM cells5'. Several other MMPs have since been identified that cleave PLG to form angiostatin K1-4. These include MMP-3 (stromelysin-1), MMP-7 (matrilysin), MMP-9 (gelatinase Bltype IV

collagenase) and macrophage-derived MMP-12 (metall~elastase)~~-~~.

Interestingly, a recent report shows that in vitro aggregated platelets, which release MMPs, also release angiostatin, though it was not determined if this generation was dependent on MMP activitys7. Macrophages have also been observed to generate angiostatin K1-4 during inflammation, in a tumor fiee setting, in a mouse models8. This formation was dependent on plasmin activity but not on MMP activitys8. Other enzymes, released by prostate cancer cells, that have been observed to generate angiostatin K1-4 include prostate specific antigen and cathepsin D ~ In addition, an as yet ~ ~ ~ ~ . unidentified 13 kDa serine protease expressed by BT325 human glioma cells was shown to liberate ~ 1 - 9 ' .

Our laboratory first showed that serum-free conditioned media collected from prostate cancer cell lines, PC-3, LNCaP and DU145, contained enzymatic activity that could convert PLG to angiostatin and that this conversion was dependent on serine protease activity''. We later showed that this occurred in a two-step reaction (Figure 2). First, uPA, secreted by the prostate cells, cleaves plasminogen to form the active serine protease plasmin, then, in the presence of a fiee sulfhydryl donor (FSD), plasmin undergoes autoproteolysis to yield angiostatin (Figure 2)47. In this system, the FSD was identified as L-cysteine in the culture media47. In tests using purified components, we determined that other plasminogen activators, tPA and streptokinase, and other FSDs (N-acetyl-L-cysteine, D-penicillamine, captopril, or reduced glutathione) could be used to generate angiostatin47.

Using mutant PLG isoforms, we also determined that this generation was dependent on active plasmin formation47. In fact, if plasmin is used as the starting substrate, a FSD is sufficient to convert plasmin to angiostatin47.

Another group has demonstrated this in vivo. In an orthotopic breast cancer model, using the MDA-MB-435 cancer cell line, N-acetyl-L-cysteine (NAC) treatment alone increased angiostatin formation and suppressed tumor We later determined that the angiostatin generated by our system

182 CYTOKTNES AND CANCER contains K1-4 and -85% of K5, with a COOH-terminus of -50% of each at

~r~~~~ Or Lys531, prompting us to refer to this isoform as angiostatin4.563.

Figure 2. Generation mechanism of angiostatin K1-4.85. A) Plasminogen is converted to plasrnin by a plasminogen activator (uPA or tPA). B) Plasmin, in the presence of a FSD (R- SH), undergoes autoproteolysis. C) Thus, angiostatin (AS) K1-4.85 is generated, consisting

of the first four of the five kringle domains of PLG plus -85% of kringle five.

Using the hingle designation, this isoform would be K1-4.85 as it contains the first four kringles plus 85% of K5. Interestingly, Aggarwal and colleagues believed that both K1-4 and K1-5 isoforms were produced in their NAC treatment studies, based on Western blot analysis (sequence analysis was not performed)62, suggesting that different tumor cell lines can

generate different isoforms. Westphal et aZ. noted that different human cancer cell types varied in their propensities toward angiostatin production, with only 217 colon and 219 renal cancer cell lines generating angiostatin while 616 bladder, 617 prostate and 24/25 melanoma cancer cell lines did64.

We had previously observed that PC-3 subcutaneous tumors in athyrnic mouse models inhibit the growth of m e t a ~ t a s e s ~ ~ , and Folkman's group had shown that PC-3 tumors secrete a circulating angiogenesis i n h i b i t ~ r ~ ~ . Thus, it is likely that K1-4.85 is the circulating factor responsible for this inhibition''. As mentioned above, we have recently shown that HT1080 fibrosarcoma and MDA-MB231 breast cancer cells can also generate angiostatin K1-4.85 and have shown that this generation can occur on the cell surface in the absence of a FSD~'. We determined that cell surface generation by PC-3 cells is dependent on the expression of cell-bound uPA and cell surface p - a ~ t i n ~ ~ . In a cell free system, p-actin can replace the FSD in angiostatin generation48. We also observed that normal fibroblasts and microvascular endothelial cells also generated K1-4.85 by this system, but at significantly lower levels than the cancer cells49. As the actin expression was similar, this was likely due to the significantly lower levels of uPA found on the surface of these normal cells49.

Similar observations were published by Stathakis et al. who observed that a serine protease, a plasmin reductase, later identified as phosphoglycerate kinase, and a FSD catalyze the proteolysis of plasmin to yield angiostatin67-69. Of interest, Stathakis and colleagues identified the same COOH-terminus of angiostatin within K5 as our own

suggesting that we have isolated the same angiostatin isoform, K1-4.85, via slightly different mechanisms. Consistent with our work and that of Stathakis, O'Mahony et al. demonstrated that conversion of PLG to angiostatin K1-4 by serum free conditioned medium from human pancreatic cancer cells was dependent upon serine protease activity in the media, i.e.

requiring plasmin activation7'. Cao et al. also reported generating angiostatin using uPA-activated plasmin72. They referred to this angiostatin as K1-5;

however, based on the reported sequence, it contains K1-4 and most of ~ 5 ' ~ , with the same N- and COOH-termini as we reported for ~1-4.8563.

The above data illustrates that cells can use different mechanisms by which to produce angiostatin isoforms. This could be merely a coincidence, a redundancy of evolution, or a means by which angiostatin production can be tightly regulated by different cell types. However, it still remains to be seen as to which isoforms and which mechanisms of generation occur in vivo and are of clinical significance. Table 1 summarizes how some of the different isoforms can be generated.

CYTOKINES AND CANCER

Table I. Isoform generation mechanisms Kringle

Domain Produced by: Reference:

K1-3 porcine elastase 10

elastase, neutrophil-secreted 73

K2-3 elastase + pepsin 74

K1-4 MMP-2 (gelatinase A) 52

MMP-3 (stromelysin 1) 55

MMP-7 (matrilysin) 53

MMP-9 (gelatinase Bltype IV collagenase) 53

MMP-12 (metalloelastase) 56

Prostate specific antigen 59

Cathepsin D 60

24 kDa endopeptidase (Chyseobacterium) 75

K1-4.85 PA + FSD 15,47

PA + p-actin (on cell surface) 48

PGK + plasmin + FSD 67-69

uPA-activated plasmin 72

~ 1 - 5 ' 13 kDa serine protease 6 1

Isolated Kringles

K1 Elastase + chymotrypsin 76

V8 protease digestion (S. aureus) of K1-3 77

K2 *

K3 *

K4 Elastase 10

K5 Elastase + pepsin 5 1

MMP-3 (stromelysin) 55

Abbreviations: PA, plasminogen activator; FSD, free sulfhydryl donor; MMP, matrix metalloproteinase; PGK, phosphoglycerate kinase. #specific COOH-terminus sequence was not determined in this study. *No enzymatic isolation schemes reported for these

kringle domains.

KRINGLE DOMAINS IN ANGIOGENESIS

As described above, the PLGIplasmin protein contains five kringle domains. Kringle domains are -80 amino acid long loop structures formed by 3 intra-loop disulfide bonds mediated by six conserved cysteine resides, with the amino acids flanking the third and fourth cysteines also highly conserved (reviewed in 35,78). Several other proteins in the hemostatic system, such as ~~rothrombin, uPA and tPA, as well as other non-hemostatic system proteins, contain kringle domains, ranging in number from just one up to the 38 or more (reviewed in 35,78). Kringle domains bind to lysine residues and serve to bind the parent protein to its substrate (reviewed in 35,78). In the case of PLGIplasmin, the kringle domains facilitate binding to the plasmin substrate fibrin (reviewed in 35,78). Some useful consequences of this binding property are that kringle-containing proteins can by isolated using lysine-sepharose columns, and their function(s) can in some cases be inhibited by lysine analogs, such as epsilon-amino caproic acid (reviewed in 79). In addition to full kringle domains, domains called short consensus repeats (SCR), often found in complement proteins, have a similar folding module, but use only two disulfide bridges (reviewed in 78). The biological function of the SCR domains is unknown, including whether or not they affect angiogenesis; however, it is known that they can mediate protein- protein interactions (reviewed in 78). The fibronectin type I1 domain, found in fibronectin, HGF, MMPs, some cellular receptors and seminal fluid proteins, also have a kringle-like structure, held together with two pairs of disulfide bridges (reviewed in 78).

As the angiostatin isoforms essentially consist of varying numbers of kringle domains, it is reasonable to hypothesize that the kringle domain structures are critical to the anti-angiogenic activity of these molecules.

Folkrnan's group and others have demonstrated that any disruption of the kringle structures in multi-kringle domains demolishes the endothelial cell inhibitory As studies proceeded with angiostatin isoforms, it became evident that differences in the number of kringle domains as well as which kringle domains were included in a given angiostatin isoform affected its activity. The amino acid sequence of each of the five kringle domains within PLGIplasmin is highly conserved, with all five being around 50%

identical to each other51. Despite this similarity, they are known to differ in their lysine binding affinities, as tested by binding to epsilon-amino caproic acid (reviewed in 79).

Based on in vitro endothelial cell assays, the anti-angiogenic activity of the different isoforms can be compared. Elastase generated angiostatin K1-4 inhibits endothelial cell proliferation with an ICjO (concentration for half maximal inhibition) of 135 n ~ ' O > ~ l . In migration studies, Ji et al. found the

186 CYTOKINES AND CANCER

1Cs0 of recombinantly generated K1-4 to be 50 nklS0. Elastase generated K1-3 has an ICS0 of 70 nM in anti-endothelial cell proliferation studiess0. In migration studies, Ji et al., found that recombinantly generated K1-3 had only marginal activity against migration with an ICso at >I000 nM80. These data suggest that the kringle components contribute different activities and that kringle 4 does not contribute to the anti-endothelial cell proliferation activity while it is critical for anti-migration activity. To test this, Ji et al.

tested a combination of K1-3 and isolated K4 against endothelial cell migration and found that this combination inhibited as well as ~ 1 - 4 8 0 . These data are consistent with studies on individual kringle domains. K1 inhibits endothelial cell proliferation with an ICsO of 320 nMsO; however, in migration studies, the IC50 is >I000 nM80. Conversely, K4 is inactive against endothelial cell proliferations0, but inhibits migration with an IC50 of 500 nM80. As Cao et al. noted, one difference between these two kringle domains is that K4 contains two clusters of positively charged lysine residues adjacent to cysteine 22 and 80 which results in an exposed and positively charged area not found in the other kringle domains7*.

Interestingly, of the isolated single kringle fragments, K5 was found to have the most potent activity against endothelial cell proliferation in vitro with an ICso of 50 nMS1. These data suggest that subtle differences in kringle sequence may provide functional specificity. Two multi-kringle domain angiostatin isoforms contain part of (-85%) or all of kringle 5, K1- 4.85 and K1-5, respectively. Our group has shown that K1-4.85 inhibits capillary endothelial cell migration with an ICSo of 0.35 pg/ml (-10 nM), and Cao et al. showed it potently inhibits endothelial cell proliferation with an IC50 of 50 P ~ 7 2 . The activities of these isoforms and isolated kringles are summarized in Table 2.

Table 2. Anti-angiogenic and anti-tumor activity of angiostatin isoforms and kringle domains In vitro inhibitory In vivo inhibitory

activity: activity:

corneal

Kringle or CAM Anti-tumor

Domain proliferation migration assay activity References

Kl-3 Yes marginal Yes Yes 8,9,80-82

K2-3 no Yes ND ND 50,80

K1-4 Yes Yes Yes Yes 8,80

K1-4.85 Yes Yes Yes Yes 15,20,72

K1-5* ND ND ND ND

K 1 Yes marginal ND ND 80

K2 Yes Yes ND ND 80

K3 Yes Yes ND ND 80

K4 no Yes ND ND 80

K5 yes yes ND yes 51,83

Abbreviations: ND, not done.

*An isoform containing K1-4 and full-length K5 has not been studied.

While the in vitro results are a valid means of comparison, the in vivo anti- cancer activity of these molecules is of most importance. K1-3, K1-4, K1- 4.85 and K5 have all been tested in mouse models of human cancers and have demonstrated anti-tumor activity against various tumor models, as listed above. However, as is well known, many molecules that work in mice do not work against cancer in humans. Therefore, final validation of the therapeutic value of these molecules awaits clinical trials. Progress in this area is discussed below.

As it became evident that the kringle domains of angiostatin were critical to the anti-angiogenic function, researchers began screening other kringle- containing proteins or kringle-containing fragments for anti-angiogenic activity. A kringle domain (K2) from prothrombin, liberated by factor Xa cleavage, was found to inhibit endothelial cell proliferation, with an ICso of 2 pglml, and inhibits neovascularization in the CAM assayg4. Naturally occurring kringle containing fragments of apolipoprotein(a) and HGF have also been found to possess anti-angiogenic a ~ t i v i t y ~ ~ , ~ ~ . In addition, kringle- containing fragments of both tPA and uPA, recombinantly expressed, have demonstrated anti-endothelial cell activity. Together, these data suggest that the kringle structure itself possesses anti-angiogenic properties; thus, studies of the kringle structure and its interaction with endothelial cells could aid in the design of novel kringle-like drugs. Sheppard et al. has taken such an approach and has generated a tetrapeptide and a dipeptide, based on an amino acid sequence within kringle 5 (lysine-leucine-tyrosine-aspartic acid), that have similar anti-endothelial cell activities as K5 in vitrog7. Likewise, Dettin and colleagues have generated linear and cyclic peptides based on the K4 sequence and several of these peptides inhibited the migration of human microvascular endothelial cellsgg. Studies such as these could refine the anti- angiogenic sequence to a small-molecule type drug that could be easily synthesized for use in the clinic.

4. ROLE OF THE HEMOSTATIC SYSTEM IN ANGIOGENESIS

Angiostatin's parent molecule, PLG, functions in the hemostatic system, which encompasses both coagulation and fibrinolysis, processes critical for wound repair and the healing process. This system is physically and functionally connected with the vasculature. Many cancer types are associated with activation of the coagulation system, i.e. are hypercoagulable (reviewed in 36). Angiostatin and PLGIplasmin are among a growing

188 CYTOKINES AND CANCER

number of proteins in the homeostatic system known to affect angiogenesis.

Other factors functioning in these pathways that have also been implicated in angiogenic regulation include tissue factor, thrombin, fibrin, plasminogen activators uPA and tPA, and activated platelets (reviewed in 36). Tissue factor, for example, in addition to activating coagulation via thrombin generation, can also increase VEGF expression by stimulating platelet activation (reviewed in 36). Interestingly, among the hemostatic system proteins involved in angiogenesis inhibition, several are cryptic fragments of larger molecules, as is angiostatin. For example, as described above, the prothrombin K2 domain is released by Factor Xa cleavageg4. Such dual functions of the parent molecule and cleavage product in hemostasis and angiogenic regulation suggest a coordinated regulation of these processes.

However, the use of cryptic fragments appears to be a common theme among angiogenesis inhibitors in general. Many other inhibitors are proteolmc cleavage products of larger molecules that are not associated with the hemostatic pathway. Table 3 lists these inhibitors and their parent molecules. It is interesting to note that many of the parent molecules are extracellular matrix components, such as collagens IVY XV, and XVIII and fibronectin. This may suggest that in the hemostatic system, as well as in other tissue environments, the release of cryptic fragments may have evolved as a negative feedback loop to regulate protease activity andlor pro- angiogenic stimuli.

Table 3. Angiogenesis inhibitors as fragments of larger molecules Generation

Inhibitor Parent Molecule Mechanism References

aaAT Antithrombin 111 thrombinielastase 89

Alphastatin Angiostatin des(Ang 1)AGT Arrestin

Fibrinogen recombinant 90

Plasminogen see Table 1 8

AGT renin 91

Collagen type IV, MMP 92

a 1 chain

Canstatin Collagen type IV, MMP 93

a 2 chain

Endostatin Collagen type XVIII elastase 94-97

FgnE Fibrinogen plasmin 98

Kininostatin HK recombinant 99

PEX MMP-2 autocatalytic 100

16 kDa Prolactin Prolactin cathepsin D 101

Prothrombin K2 Prothrombin factor Xa 84

Restin Collagen XV recombinant 102

Tumstatin Collagen type IV, MMP 103-105

a 3 chain

Vasostatin Calreticulin not known 106,107

Abbreviations: aaAT 111, anti-angiogenic antithrombin; AGT, angiotensinogen;

FgnE, Fibrinogen E fragment; HK, high molecular weight kininogen; MMP, matrix metalloprotease.

5. MECHANISM(S) OF ANGIOSTATIN ANTI- ENDOTHELIAL CELL ACTIVITY

Many different cell types are capable of generating angiostatin, either through secretion of enzymes or the presence of cell surface molecules.

However, the angiostatin isoforms have shown high specificity for action on endothelial cells, with only a few exceptions noted thus far. The first exception was noted by Walter and Sane. They found that angiostatin inhibited HGF-induced proliferation and migration of smooth muscle cells in vitro and co-localized with these cells in v i v ~ ' ~ ~ . Angiostatin isoforrns K1-4 and K1-3 have also been observed to inhibit neutrophil migration induced by several chemokines and by the HIV-Tat protein in an in vitro chemotaxis assay and in an in vivo Matrigel implant m ~ d e l ' ~ ~ * " ~ . Another study suggests that it may also inhibit osteoclast activity, thus inhibiting bone resorption"'. However, most studies have focused on its endothelial cell activity. As discussed above, angiostatins can inhibit endothelial cell proliferation, migration, and tube formation. The mechanism of this inhibition, however, is still under investigation.

A few studies have suggested that angiostatin inhibits endothelial cells by blocking specific inducer-mediated activities or by modulating angiogenic regulator expression by endothelial cells themselves. Our group reported that angiostatin K1-4.85 could inhibit PLGIplasmin-enhanced in vitro invasion of endothelial cells and melanoma cells that express PA^^. ^{AS we}

also observed inhibition of tPA-catalyzed plasminogen activation, we hypothesized that this could be a mechanism of this inhibition as we also noted high affinity direct binding of K1-4.85 to PA^^. Another group suggests that angiostatin acts by specifically blocking HGF-induced activity of endothelial cells"*. Hajitou et al. observed down-regulation of VEGF expression following angiostatin treatment19, and another group has shown that angiostatin reduces VEGF- and bFGF-induced activation of mitogen activated protein kinases, ERK-1 and -2'13. In retinal capillary cells, angiostatin treatment both down-regulated VEGF and up-regulated pigment epithelium-derived factor, another potent angiogenesis inhibitor'14. These studies suggest that angiostatin functions by blocking inducer activity.

On the other hand, many studies have demonstrated that angiostatin treatment triggers apoptosis of endothelial cells. We showed that K1-3, K1- 4 and K1-4.85 all induce apoptosis of endothelial Folkman's group also showed that angiostatin K1-3 induced apoptosis and further showed that

190 CYTOKINES AND CANCER angiostatin treatment induced activity of focal adhesion kinase by an RGD- independent ~ a t h w a ~ " . Subsequently, Lu et al. showed that K5 induced endothelial cell apoptosis as we11116. Our group went on to show that induction of apoptosis by angiostatin K1-4.85 involved activation of caspases-8, -3 and -9117. Moser et al. demonstrated angiostatin K1-3 binding to the alp-subunits of ATP synthase on the surface of human umbilical vein endothelial cells, and this binding was not inhibited by an excess of PLG indicating it was a unique binding site for angiostatin16. Though ATP synthase is usually contained within the cytoplasm, another group had previously demonstrated its expression on the surface of tumor cells"*. Cao and colleagues also demonstrated that angiostatin bound to the alp-subunits of ATP synthase and that this binding induced apoptosis1'9. Again, this isoform was referred to as K1-5; however, based on the referenced sequence7', it is K1-4.85.

Consistent with our previous study, they showed that the apoptosis induction involved the sequential activation of caspases-8, -9 and -3'l9.

Furthermore, they demonstrated endothelial cell apoptosis in vivo in a fibrosarcoma tumor model where co-administration of angiostatin with a caspase-3 inhibitor blocked endothelial cell apoptotic induction119. An interesting theory is that the low pH environment of the tumor promotes translocation of ATP synthase to the cell surface in caveolae, and subsequent angiostatin binding causes a precipitous drop in intracellular pH, thus triggering apoptosis120. Another group demonstrated that angiostatin treatment induced p53-, Bax- and tBid-mediated cytochrome c release and activated the Fas pathway'21. Weichselbaum's group suggest that angiostatin's pro-apoptotic induction is mediated by the sphingolipid second messenger ceramide and RhoA activation1", and Sharma and colleagues show an association with down-regulation of the cyclin-dependent kinase

5123

The majority of the apoptosis studies, with the exception of the Moser et al. and Veitonmaki et al. studies, did not identify the endothelial cell surface receptor for angiostatin. While ATP synthase has been linked to apoptotic induction, other potential receptors have been identified. These include annexin 11, angiomotin, and a,P3. However, the role of these candidate receptors in angiostatin activity is less clear. Tarui et al. found that angiostatins K1-3, K1-4 an K1-5 bind to avP3 on the surface of bovine arterial endothelial cells124. They later demonstrated that angiostatin may act by blocking plasmin-induced activity via blocking plasmin binding to

avp344,124 . Using a yeast two-hybrid library, angiostatin was also found to bind to angiomotin, a protein localizing to the leading edge of migrating endothelial cell^''^. This group suggests that angiomotin promotes

angiogenesis, and angiostatin binding inhibits this Sharma and colleagues demonstrated binding to annexin I1 on bovine aortic endothelial cells12'; however, a mechanism of action for this receptor is still unknown.

Overall, the above data suggests that angiostatin may have multiple effects on endothelial cells and angiogenesis, at least in vitro. Differences in cell types used (species as well as vessel type) and in assays used could significantly influence results. The in vivo data, however, supports an apoptosis induction model of angiostatin activity, although additional studies are needed to elucidate the mechanism of this activity.

6. ANGIOSTATIN ISOFORMS IN WVO

As discussed above, many different angiostatin isoforms can be generated under laboratory conditions; however, it is of interest to determine which are produced in vivo under normal and/or pathophysiologic conditions. While angiostatin was initially isolated from tumor-bearing mice, several studies indicate that angiostatin isoforms are also present in human tissues and fluids. In a small pilot study, our group observed only the angiostatin K1-4.85 isoform in normal human plasma, at approximately 6 to 12 n ~ This level was also observed in the plasma collected from cancer ~ ~ . patients63. However, we observed markedly higher levels of K1-4.85 in ascites from ovarian cancer patients as well as in ascites from patients with nonmalignant etio~ogies~~. Another group confirmed this observation. They observed a 55 kDa angiostatin isoform in the ascites of patients with abdominal cancers12'. Several other studies have identified angiostatin isoforms in urine. Sten-Linder et al. compared angiostatin levels in urine between 117 cancer patients and controls by densitometry of Western blots and found that the levels in cancer patients (27k75 pg/L) were significantly higher than in the normal controls (353 p g ~ ) 1 2 9 . In their study, multiple immunoreactive bands were detected using an anti-K1-3 antibody129. Cao and colleagues also measured angiostatin and PLGIplasmin in urine130. They found low levels of PLGIplasmin and no detectable angiostatin in the normal controls compared to high levels of PLGIplasmin and detectable angiostatin in the cancer patients130, though the exact isoform was not determined.

It is important to bear in mind that the regulation of angiogenesis is important in settings other than cancer. Sack et al. studied the diurnal variations in angiostatin levels in human tear fluid131. While open eye tear fluid from all normal individuals contained low levels of PLG and no detectable angiostatin, tears collected after overnight eye closure contained significant amounts of PLG, and various angiostatin-related isoforms, including K1-3, K1-4, and possibly isolated K5. They hypothesize that these

192 CYTOKINES AND CANCER angiostatin isoforms may play a role in preventing neovascularization in the hypoxic environment of the closed eyel3l. From all of the above observations, it appears that angiostatin isoforms play a significant role in normal physiologic and pathophysiologic functions.

PROGRESS IN ANGIOSTATIN THERAPIES

Developing angiostatin isoforms for use clinically is the ultimate goal of angiostatin research. It has great potential use not only for cancer treatment, but also for the treatment of other diseases, which involve inappropriate angiogenesis, including diabetes, retinopathies, and rheumatoid arthritis. For cancer treatments, anti-angiogenic therapies offer many advantages over traditional therapies in principal. Firstly, they are not mutagenic so secondary tumors are unlikely. Secondly, they target the tumor vasculature specifically; therefore, there are fewer side effects. Thirdly, they can act synergistically with current chemo-, radiation- and gene-therapies132-136.

Fourthly, as they do not target the tumor cell, but instead the genetically stable vascular endothelial cells, the tumor cells are unlikely to develop resistance to therapy'37"38. Lastly, they can be easily delivered via the circulation to their target cells (reviewed in 2,139). Unfortunately, to date, most clinical trials with angiogenesis inhibitors alone have not seen significant tumor regression; however, trials combining these agents with cytotoxic chemo-therapies have shown promising results (reviewed in 140).

One disadvantage of anti-angiogenic therapies lies within one of the advantages. As they do target the genetically stable vascular endothelial cells, and not the tumor cells, they do not kill the tumor cells and are therefore not curative. However, viewing anti-angiogenic therapy as a long- term maintenance treatment or control treatment and/or combining these treatments with traditional therapies easily compensates for this deficit.

Angiostatin has shown synergy with other treatment modalities. In collaboration with Mauceri and colleagues, we validated the theory that

141,142

angiostatin could potentiate radiation therapy of cancer in mice .

Additional studies have since confirmed these results and shown synergy

143,144

with other chemotherapies . These studies indicate that combining angiostatin with traditional cytotoxic therapies such as radiation therapy andlor chemotherapy may prove to be effective clinically. However, the most effective dosing schedules for anti-angiogenic agents differ significantly from traditional cancer therapies. Traditional therapies are given on a maximum tolerated dose regimen, with high dose, short-term treatment scheduling with extended breaks between treatments to allow

recovery of normal tissues (reviewed in 145). In animal models, chemotherapeutic drugs, such as vinblastine and cyclophosphamide, when used to target the tumor vasculature, were found to be more efficacious when administered on a frequent, low dose schedule rather than on a maximum tolerated dose s ~ h e d u l e ' ~ ~ - ' ~ ~ . This low, frequent, or

"metronomic," dosing increases the efficacy against the tumor-associated vascular endothelial cells and greatly diminishes toxicity in animal models149. How efficacious it is in patients, however, remains to be seen.

Whether angiostatin is used alone or in combination with traditional therapies, the pertinent question is still the same: what is the most efficacious method of delivering angiostatin to the patient? There are several options, the most direct of which is to administer angiostatin protein directly to the patient. Purified protein could be generated in two ways. One is the purification of angiostatin from cleavage of human PLG isolated from plasma. However, producing active angiostatin in large quantities by this strategy has proven to be technically difficult and labor intensive and has potential for contamination from plasma derived pathogens. Thus, this method is not practical for large-scale production. The second option would be to produce recombinant human angiostatin. In the initial effort, Sim et al.

used the Pichia pastoris expression system150. Recombinant angiostatin K1- 3 has been demonstrated to be biologically active and inhibited tumors in

mice150,151 . Recombinant K1-3 and K1-4 proteins, also generated in the Pichia pastoris expression system, suppressed B 16-BL6 lung metastases by greater than 80% when administered at 30 n ~ / k g / d a ~ ' ~ ' . However, as Pichia pastoris do not express proteins containing kringle domains, it is not clear if the post-translational processes required for proper kringle assembly are in place in this system. The degree of correct kringle conformation has not been studied, nor is it known how this would affect activity in cancer patients. As an alternative strategy, Meneses et al. used a mammalian expression system152. This mammalian-derived angiostatin K1-3 suppressed intracranial brain tumor growth in immune-competent rats up to 85%, with a 32% decrease in tumor neovasc~larization~~~. This is encouraging, however, the in vivo half life of angiostatins and K5 are very short, -15 minutes (reviewed in 120). Therefore, to achieve therapeutic levels, high doses are needed. In addition, due to the nature of anti-angiogenic therapy, i.e.

limiting tumor growth rather than killing the tumor, chronic administration is also needed. Thus, large quantities of purified protein would be required.

These factors would make the use of recombinantly produced angiostatin an expensive form of therapy.

Despite these obstacles, clinical trials have been progressing using recombinant angiostatin K1-3 (rK1-3), produced by Entremed. Two phase 1 trials have been conducted at the Kirnmel Cancer Center at Thomas

194 CYTOKINES AND CANCER

Jefferson University in Philadelphia. The first, initiated in the year 2000, was to evaluate safety of rK1-3 doses in patients with cancer. Twenty-four patients with various solid tumors were enrolled with five patients receiving the therapy for more than a year153. Dosing was at 7.5 to 30 mg/m2/day divided into two injections dailylS3. Complications observed included two patients who developed hemorrhage in brain metastases and two who developed deep vein t h r o m b ~ s i s ' ~ ~ . No abnormalities in coagulation parameters were notedlS3. Thus, it appears that long-term therapy with rK1- 3 is well tolerated. A second phase 1 trial was initiated to evaluate the combination of rK1-3 and radiation therapy in cancer patients with solid tumors, and currently, a phase I1 trial has been initiated to test rK1-3 in combination with paclitaxel and carboplatin treatment in patients with non- small-cell lung cancer, again at Thomas Jefferson University. The phase I1 trial is currently enrolling patients. Entremed, however, has recently ceased development and testing of rK1-3 and has turned over the rights of rK1-3 to Children's Medical Center Corporation (announced Feb. 2,2004).

Another possible approach to delivering angiostatin to the patient is through the use of gene therapy. Expression of angiostatin can be induced in cancer cells by in vitro transfection of an angiostatin-expression vector plasmid, or by use of recombinant retroviral or adeno-associated virus vectors that carry genes coding for angiostatin'4'221'54"55. Implantation of angiostatin-expressing cells into mice resulted in reduced tumor growth and tumor-associated angiogenesis. More recently, intratumoral injection of liposomes complexed to plasmids encoding angiostatin reduced the size of tumors implanted in the mammary fat pad of nude mice by 36%30. The use of gene therapy is an attractive model to induce angiostatin expression and suppress tumor growth; however, it is subject to the broader concerns about gene therapy in general, specifically whether high levels of expression can be achieved and the safety of the approach.

One of angiostatin's interesting properties as an angiogenesis inhibitor, as a fragment of a larger protein, gives rise to another option for the delivery of angiostatin to the patient. As we and others have shown, angiostatin can be produced by proteolytic cleavage of PLG (see above). Therefore, one could develop therapies in which angiostatin is generated in vivo from the PLG present in the patient's own blood. As discussed above, we have demonstrated that PLG is converted to K1-4.85 in a two-step reaction requiring a plasminogen activator and a F S D ~ ~ . Thus, by administering a plasminogen activator and a FSD, an "angiostatic cocktail" treatment, to a patient, angiostatin could be generated in vivo. This treatment modality is particularly attractive because plasminogen activators (uPA, tPA and streptokinase) and FSD (Captopril, n-acetyl-L-cysteine, Mesna and D-

penicillamine) are clinically available. Interestingly, Captopril alone was shown to inhibit angiogenesis in ratslS6, and one patient with Kaposi sarcoma showed stable disease over a six month period on captopril after failing chemo- and radiation therapieslS7. However, in these studies, angiostatin generation was not investigated as the association with FSD was

just coming to light. -

Day 1 Day 2 Day 3 1 2 3 4 5 6 7 8 9

Complexed

3

+

Figure 3. Angiostatic cocktail treatment generated angiostatin K1-4.85 in a patient with colon cancer. The Angiostatic Cocktail (tPA + captopril) was administered for five consecutive days (days 1-3 illustrated here). On each day, prior to administration of the Angiostatic Cocktail (9:OO; lanes 1,4,7), the angiostatin K1-4.85 levels were undetectable ( 4 0 nM). During tPA infusion (10:30; lanes 2,5,8) and at completion of infusion (15:30;

lanes 3,6,9), angiostatin K1-4.85 levels increased to 100 nM and the detectable plasminogen is reduced. Additional data indicate that the large (>I00 kD) bands on western blot which

cross-react to anti-angiostatin K1-4.85 antibodies, are indeed a series of compiexes of angiostatin K1-4.85 with as yet undefined other proteins (Soff, unpublished observations).

An initial pilot study experience with seven patients, presented at the American Association of Cancer Research meeting in 2000, validated this theory. The angiostatic cocktail treatment induced angiostatin K1-4.85 formation and antiangiogenic activity was induced in plasma63. In this study, several patients had some tumor regression, including one with a complete remission63. A Western blot showing induction of K1-4.85 formation in a patient with extensive, metastatic, refractory colon cancer is shown in Figure 3. Currently, a phase I trial is underway at Northwestern University Feinberg School of Medicine (Chicago, IL), testing tPA in